Fig. 1: Workflow from data collection, DNA extraction, taxa identification, towards the quantification of extirpation and potential extinction.

We collected eight lake sediment cores from seven sites in Siberia and Alaska. DNA was extracted from subsamples taken at different depths, corresponding to various sediment age layers. After amplifying the tnrL P6loop region, amplicon sequence variants (ASVs) were identified and assigned to taxa using the SibAla_2023 database, with confidence levels ranging from 90% to 100%. To discern true taxonomic signals from potential chimeric sequences caused by PCR or sequencing errors, we developed a co-occurrence-based community detection approach. This analysis identified 359 taxa, with 216 taxa matching the SibAla_2023 database at 100% confidence (dbtaxa) and 143 taxa matching at 90–99% confidence (non-dbtaxa). From these 359 Taxa, 242 taxa are also present in the modern time-slices and are thus identified as neither extirpated nor extinct. Of these 359 taxa, 242 are also present in modern time-slices, indicating they are neither extirpated nor extinct. Among the remaining 117 taxa, 67 are considered extirpated, while 60 are absent from the region-specific SibAla_2023 database. These 60 taxa could be missing from the database, extirpated from Siberia and Alaska, or potentially globally extinct. After evaluating database coverage with a synthetic dataset simulated using data from GBIF, we found that 42 taxa could be missing due to insufficient coverage in the SibAla_2023 reference database. This suggests that 18 (60 minus 42) non-dbtaxa might be globally extinct. Further checks with global databases showed that 54 of the 60 taxa (88%) could not be assigned. Among the six taxa matching global databases, there is an 18% likelihood of having refugia outside Siberia and Alaska, suggesting that one taxon (1 out of 18–60) might only be extirpated. In conclusion, 17 taxa found in sediments from the last 28,000 years are potentially globally extinct. Modified DNA strand Vectorportal.com (https://vectorportal.com/) by CC BY (https://creativecommons.org/licenses/by/4.0/).